paper-summarize

Pass

Audited by Gen Agent Trust Hub on Jul 13, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted paper abstracts, which creates an indirect prompt injection surface.
  • Ingestion points: Abstract content is read from markdown files located in raw/<cluster>/<paper-slug>.md.
  • Boundary markers: The skill instructs the LLM to anchor findings to the abstract text and utilizes JSON-validated output formats.
  • Capability inventory: The skill performs file system writes to an Obsidian vault, modifies Zotero child notes, and executes shell commands through the research-hub CLI.
  • Sanitization: While output is JSON-validated, there is no explicit sanitization defined for the input abstract text before it is sent to the LLM.
  • [COMMAND_EXECUTION]: The skill utilizes shell commands to run the research-hub utility and various LLM CLI adapters (e.g., claude, gemini). It also includes documentation for running test suites using pytest. These are considered vendor-specific operations within the skill's intended environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 13, 2026, 12:52 AM
Security Audit — agent-trust-hub — paper-summarize